Agentic AI Foundation · Oct 24, 2026 · 10:18 AM

Alibaba hands OpenSandbox, its self-hosted sandbox runtime for AI agents, to the AAIF

Announced Oct 22, the Apache-2.0 project (about 16K GitHub stars) gives coding, GUI and browser agents isolated environments to run commands, edit files and execute code on Docker locally or Kubernetes at scale, with SDKs for Python, Java/Kotlin, TypeScript, .NET and Go plus an MCP server. Its sandbox protocol defines lifecycle and execution APIs so teams can plug in custom runtimes, and the repo now lives at opensandbox-group/OpenSandbox.

Read at aaif.io

More news

View all

Hugging Face hands OpenEnv, its framework for agent RL training environments, to the AAIF

Announced Oct 22, OpenEnv packages RL environments as containerized services that expose an agent interface, MCP, and the Gym protocol training libraries use, so one environment can plug into different trainers and harnesses. It's run by a multi company technical committee (Hugging Face, Meta PyTorch, Microsoft, NVIDIA and others), and next on the roadmap are environment validation and training recipes (AAIF blog)…

Agentic AI Foundation 37 minutes ago

Replit Agent can now turn your workflows into reusable workspace skills, with your approval

In the Oct 23 update, Agent can suggest new skills or improvements to existing ones based on your chats and projects, and you review and approve each before it lands in the workspace skills library. Eligible top up customers can also keep published apps running at zero credits up to an overage limit, while paid Agent pauses and Free mode stays available…

Replit Changelog 3 hours ago

OpenSpec 1.15 ships standalone binaries, so the spec-driven workflow no longer needs Node.js

Released Oct 24, every OpenSpec release now ships macOS, Linux (glibc and musl) and Windows binaries for x64 and arm64 with SHA256SUMS , installable with a one line curl ... | sh script, which helps in CI runners and agent sandboxes without Node. It also adds openspec update dry run (with json ) to preview file changes before writing anything…

GitHub (Fission-AI/OpenSpec) 4 hours ago

OpenAI hands Open Responses, its multi-provider LLM API spec, to the AAIF with AWS backing

Announced Oct 22, the Responses API based spec now sits under the Agentic AI Foundation next to MCP and A2A, so one request and streaming format can target any provider that implements it; AWS employs a core maintainer, and Vercel AI Gateway and Databricks already support it. On Bedrock you can point the OpenAI SDK at https://bedrock runtime.{region}.amazonaws.com/openai/v1 to call Anthropic, Meta, Mistral, DeepSeek and other models (The New Stack)…

AWS Open Source Blog 5 hours ago

Moonshot launches Kimi K3.1 Pro, a 1M-context open-weight MoE, with weights due by Nov 4

Launched Oct 23, the 2.8T total/104B active multimodal model is live on Kimi.ai, Kimi Code, and the API ( kimi k3.1 pro ), plus a Highspeed variant that runs 2–3x faster at about 2x the Kimi Code quota. In Kimi Code it needs a Plus plan (256K context) or Pro for the full 1M, and thinking can't be turned off (Kimi Code models)…

Kimi Blog 6 hours ago

Spotted something we missed? Start a thread.